# 6thSense

> 6thSense builds touch-aware capture hardware for robot learning: Skin, a 464-channel wearable tactile sensor, and Eye2 Pro, an egocentric stereo camera.

6thSense makes tactile and egocentric capture hardware for robot learning. Skin records what a human hand feels while it works; Eye2 Pro records what the wearer sees. Together they capture touch-aware human demonstrations for training dexterous robots.

## Products

- [Skin](https://6thsense.dev/products#skin): Wearable tactile skin. Every touch, recorded as force.
  - Tactile array: 464 sensing channels
  - Coverage: Fingertips, finger pads, and palm
  - Force resolution: 12-bit (~0.1N)
  - Motion sensing: 9-axis IMU (accelerometer, gyroscope, magnetometer)
  - Data rate: 200 Hz USB and Wi-Fi
  - Hand tracking: AprilTag fiducial markers for optical pose tracking
  - Variants: Left and right hand
- [Eye2 Pro](https://6thsense.dev/products#eye2): The egocentric camera that sees what the hand feels. First-person video, synced to touch.
  - Sensors: Dual global-shutter stereo cameras
  - Resolution: 1920 × 1200 per eye
  - Frame rate: 30 fps
  - Field of view: 132° diagonal / 121° horizontal / 92° vertical
  - Stereo baseline: 80 mm, calibrated per unit
  - Inertial data: Accelerometer and gyroscope at ~300 Hz, embedded frame-synced in the video
  - Storage: Removable microSD
  - Connectivity: Wi-Fi (hotspot or site network), USB-C, automatic cloud upload
  - Controls: One-button recording, status light, browser and companion app

## Glossary

- Tactile sensing: Measuring touch: where something makes contact and how hard it presses. Robots use tactile data to grip without crushing or dropping what they hold.
- Egocentric video: Video recorded from the wearer's own point of view, with a head-worn camera that sees what the person sees while they work with their hands.
- Tactile array: A grid of individual touch sensors. Each sensing channel reports pressure at one spot, so 464 channels together map where and how hard the hand is touching.
- Force resolution: The smallest change in force a sensor can report. 12-bit means 4,096 steps per reading; 0.1 N is roughly the weight of a 10-gram object.
- IMU (inertial measurement unit): A motion sensor. The accelerometer and gyroscope track movement and rotation; a 9-axis IMU adds a magnetometer, a compass for heading.
- AprilTag: A printed square marker, like a simple QR code, that a camera can locate precisely. It lets a camera track the hand's position and orientation.
- Global-shutter stereo camera: Two cameras side by side, like a pair of eyes. Global shutter captures every pixel at the same instant, so fast hand motion is not skewed.
- Field of view: How much of the scene the camera captures, given as the angle across the image's diagonal, width and height.
- Stereo baseline: The distance between a stereo camera's two lenses. Comparing the two views gives depth. Calibrated per unit means each camera's exact geometry is measured individually.
- Frame-synced inertial data: Motion readings stored inside the video itself, so every frame carries its matching accelerometer and gyroscope data with no separate clock to line up.

## Company

- [About](https://6thsense.dev/about): founders Alex Hyungwoo Noh, Matt Wulff, James Baek, Ronak Agarwal.
- [Careers](https://6thsense.dev/careers)
- [Contact](https://6thsense.dev/#contact): the enquiry form for Skin, Eye2 Pro, demos and data.
- Supported by people at Y Combinator, Robinhood, NVIDIA Inception Program, Tesla, Great Wave Ventures, The University of Chicago, Georgia Tech.